A7.B. The Effects of Personality and Cognitive Measures...

Almlund, Duckworth, Heckman, and Kautz 11/13/2010 75

A7.B. The Effects of Personality and Cognitive Measures on Crime and Deviance by Amanda Agan

There is a large literature in criminology focusing on the effects of self-control on crime

due to Gottfredson and Hirschi’s seminal work in A General Theory of Crime (1990). In it they

posit that a uni-dimensional factor they label as “self-control” is responsible for much of the

variance in crime and deviance across individuals – although opportunity to commit crime

interacts with self-control in important ways. If one has both opportunity and low self-control

then crime is very likely, but without opportunity no crime occurs. They also argue that self-

control is a stable trait that is unchanging over time, and thus advocate for cross-sectional tests of

the theory, since longitudinal tests would be unnecessary (and more costly).

People with low self-control, as characterized by Gottfredson and Hirschi [1990], are

“impulsive, insensitive, physical (as opposed to mental), risk-taking, short-sighted, and non-

verbal” (1990:90). Criminal acts, they argue, are “likely to be engaged in by individuals

unusually sensitive to immediate pleasure and insensitive to long-term consequences” (1990:2).

Thus parts of Gottfredson and Hirshi’s definition of self-control resembles the economists

measure of the discount rate as well as the psychologists measure of impulsivity. There have

been many studies in criminology and psychology that look at the correlations between self-

control or impulsivity and criminal outcomes and deviance. However, there have been no

studies linking common economic experimental measures of discount rates (or risk aversion) to

criminal outcomes. This seems surprising particularly given that proliferation of research

recently using these measures to predict outcomes such as education and migration (i.e. Jaeger,

Dohmen, Falk et al. [2010]).

Stephanie

Text Box

From the Web Appendix for The Relevance of Personality Psychology for Economics. Draft, November 2010 -- Do Not Cite.


Two empirical measures of self-control have emerged in the literature. The first is a

series of Likert-scale questions devised by Grasmick, Tittle, Bursik et al. [1993] – their original

factor analysis on these questions determined that, as Gottfredson and Hirschi posited, this trait is

unidimensional. The other empirical measure asks behavioral questions (Tittle, Ward and

Grasmick [2003]) – these consists of questions about 10 forms of problem behavior such as

smoking, drinking, overeating, using seatbelts etc..), and Benda [2005] includes questions like

have you skipped school, have you had sex with more than one person without a condom, have

you harassed someone in the past year, have you received a ticket for reckless driving, etc… The

behavioral measures have been criticized as verging on tautological due to the fact that they

essentially ask about participation in deviant behavior then use those responses to predict deviant

behavior.

Given that the “analytic” self-control scales consist of Likert-type personality questions it

is interesting to ask how it overlaps with the Big Five Personality traits. O’Gorman and Baxter

[2002] test how the Grasmick, Tittle, Bursik et al. [1993] index correlates with the

Conscientiousness scale of the NEO – which contains six subscales: dutifulness, self-discipline,

deliberation, competence, achievement striving, and order. They find that low self-control is

significantly negatively correlated with each of these subscales, with correlation coefficients

ranging between 0.38 (competence) to 0.57 (order). Interestingly, when they regress self-

reported deviant behavior on self-control and conscientiousness they find that self-control does

not significantly add to the prediction once conscientiousness was entered.

Both the “analytic” and “behavioral” measures, and measures like them, have proven

over and over again to consistently predict crime and deviance. Pratt and Cullen [2000]

performed a meta-analysis of 21 empirical studies that consider the effects of self-control on


crime. They find consistently positive and significant standardized correlation coefficients

between self-control measures and crime or deviance, though the correlations sizes are higher

when a behavioral measure is used. Benda [2005] finds similarly that in an OLS regression of

person or property offenses committed on low self-control that self-control significantly predicts

offenses and this effect is higher when a behavioral rather than Grasmick, Tittle, Bursik et al.

[1993] scale is used. Vazsonyi, Pickering, Junger et al. [2001], using the International Study of

Adolescent Development (ISAD) finds that the Grasmick, Tittle, Bursik et al. [1993] scale

explains a positive and significant amount of the variance in deviant acts (ranging from 10-16%

depending on the type of act) across the four countries in the study (Hungary, the Netherlands,

Switzerland, and the United States).

However, that is not to say that Gottfredson and Hirschi’s theory is completely accepted.

Although the main thrust of the theory is that self-control predicts much of the variance in

criminal acts among people there are other facets of the theory that have not held up as robustly

to empirical study. One is that once self-control is accounted for very little else should help

predict crime differences. However, Pratt and Cullen [2000] find that across several studies that

add in social learning variables the mean effect size of self-control on crime was unchanged but

the social learning variables also had significant effects on crime. Gottfredson and Hirschi also

argue that the self-control factor is uni-dimensional, however Tittle, Ward and Grasmick [2003]

among others find evidence of multi-dimensionality in the scales measuring self-control.

In addition to self control other realms of personality or non-cognitive skills have been

measured and analyzed in an attempt to determine their effects on crime. The studies mostly

analyze deviance in children/adolescents and crime outcomes of undergraduates. Two of the

studies use personality measures that load onto three main factors – constraint, positive


emotionality and negative emotionality. Across two studies – the Pittsburgh Youth Survey and

the Dunedin Multidisciplinary Health and Development study – constraint was negatively

correlated with self-reported delinquency while negative emotionality was positively correlated.

Neither study found an effect of positive emotionality on delinquency. (Caspi, Moffit, Silva et

al. [1994]). Similarly, Agnew, Brezina, Wright et al. [2002] found that their single factor

measure of negative emotionality/constraint was significantly, positively associated with self-

reports of delinquency amongst 12-16 year olds in the 2nd wave of the National Survey of

Children. Sensation seeking and impulsivity have also been found to be positively associated

with crime (Horvarth and Zuckerman [1993]), although one may argue that these traits are a

subset of self-control and thus may also offer evidence in favor of Gottfredson and Hirschi’s

theory.

These studies in criminology and psychology, for the most part, use either correlations or

OLS regression to talk about “effects” of various personality measures on crime. There has been

little attempt to rigorously deal with causation in these studies.

In addition there is an emerging literature on the effects of education on crime. Lochner

and Moretti [2004]using US data and Machin, Marie and Vujic [2010] using UK data find that

increasing years of schooling is negatively associated with crime participation. Both papers use

compulsory schooling laws as an instrument and find that these results hold up in a two stage

least squares (2SLS) analysis. Lochner and Moretti [2004] posited several mechanisms through

which education could causally impact criminal activity, including that education may change

traits such as patience or risk aversion. Though neither paper attempts to differentiate amongst

the various mechanism through which education could effect crime, that education changes ones

non-cognitive skills seems like a particularly plausible argument. This is especially true in light


of findings by Cunha, Heckman and Schennach [2010]that criminal participation is more heavily

loaded on noncognitive skills, thus any intervention that can change these (in a “positive” way) is

likely to have effects on crime.


Table A13. The Effect of Personality on Crime33

Main Variable(s) Data and Methods Causal Evidence Main Result(s)

Caspi, Moffit, Silva et al. [1994]

Outcome(s): Delinquency -- self-reports, teacher and parental reports, and official records Explanatory Variable(s): personality -- “Common language” version of California Child Q-sort (CCQ) for United States and Multidimensional Personality Questionnaire (MPQ) for New Zealand

Data: US: 430 12- and 13- year old boys from the Pittsburgh Youth Survey (PYS) NZ: Dunedin Multidisciplinary Health and Development Study (New Zealand) -862 18 year olds who took the MPQ Methods: correlations

Controls: US: Separate correlations computed for blacks and white, no controls NZ: separate correlations for men and women, but no controls Timing of Measurements: US: Reports were contemporaneous with test and were about delinquency at age 12 and 13. NZ: Self-reports are contemporaneous with MPQ test at age 18, though may have occurred in any of the previous years Theory: Personality can affect delinquent behavior

Both personality inventories were used to construct three “superfactors” – constraint (combines traditionalism, harm avoidance and control), negative emotionality (combines aggression, alienation, and stress reaction), and positive emotionality (combines achievement, social potency, well-being and social closeness) . US: Constraint is negatively correlated with self-reported delinquency for blacks and white (-0.17 and -.022 , p<0.05). Negative emotionality is positively correlated with self-reported delinquency for blacks and white (0.13 and 0.20, p<0.05). Positive emotionality is not significantly correlated with self-reported delinquency. Similar signs and significant for teacher- and parent-reported delinquency, except for positive emotionality which is negatively correlated with parent-reported delinquency for blacks and whites (-0.26 and -0.21 (p<0.05) NZ: Constraint is negatively correlated with self-reported crime for men and women (-0.44 (p<.05)), negative emotionality is positively correlated with self reported crime for men and women (0.48 and 0.34 (p<.05)), positive emotionality is not significantly correlated with self-reported crime. Signs of correlations were consistent across informant reports, police contact and court convictions, and for the most part similarly significant. Overall MPQ profile can explain 34% (25%) of variance in self-reported criminal activity for men (women).

33 Table A13 - Table A15 were created by Amanda Agan.


Horvarth and Zuckerman [1993]

Outcome(s): Crime -- self reports of arrests for drugs, shoplifting, DUI, perjury, forgery, or vandalism Explanatory Variable(s): Sensation Seeking -- form V of the Sensation Seeking Scale (SSS). Impulsivity -- Narrow Impulsivity Scale of Eysenck and Eysenck (1978)

Data: Collected by authors; 447 undergraduates from an Introduction to Psychology course at the University of Delaware Methods: Correlations and multiple regression

Controls: Perceived proportion of peers participating in the specific criminal behavior, perceived risk of negative consequence from the specific criminal behavior Timing of Measurements: Personality measures and questions about crime were contemporaneous, though crime questions asked about criminal activity ever Theory: Sensation seeking, impulsivity and peer behaviors are likely important influences on criminal (and other risky) behavior

Sensation seeking and impulsivity are positively correlated with self-reported crime risk factor (0.53 and 0.36 respectively (p<0.01). In a multiple regression analysis with both sensation seeking and impulsivity included as well as the other controls, both sensation seeking and impulsivity are still positively associated with self-reported crime (with “Beta weights” of 0.27 and 0.13 - no p-values given but stated significant in the text). Note that perceived behavior by peers was the strongest predictor of criminal behavior.


Agnew, Brezina, Wright et al. [2002]

Outcome(s): Juvenile Delinquency -- self-reported five-item scale of how many times they had committed 5 acts of delinquency (hurt someone, stolen from store, damaged school property, skipped school, gotten drunk) Explanatory Variable(s): Negative emotionality/constraint -- factor derived from questions from teachers and parents about the child’s behavior and personality

Data: 2nd wave of the National Survey of Children - 1423 children who completed the interviews, age 12-16 in 1981. Methods: Correlations and OLS

Controls: sociodemographic characteristics: total family income (8 categories), education of primary parent, family status (divorced, married), age of child, sex of child, race of child; measures of “strain” - including conflict with parents, school hatred, picked on by kids, family strain and neighborhood strain; measures of social control and social learning such - attachment to parents, parental firmness, school commitment, educational goals, time spend on homework, school attachment, troublesome friends Timing of Measurements: Contemporaneous self-reports of crime with questions to parents/teachers about behavior, although both could describe actions in the past Theory: Experiencing strain can increase an individuals likelihood of experiencing negative emotions which in turn create pressure to take action which may take the form of delinquency/crime. Personality traits, such as negative emotionality, may have a significant impact on this link.

Negative emotionality/low constraint is positively correlated with delinquency (0.22, p<0.01). In a regression of delinquency on all controls and negative emotionality/low constraint, an increase in the negative emotionality/low constraint scale increases delinquency (p<0.01). It is unclear how to interpret this magnitude. When negative emotionality/low constraint is interacted with strain, the main effect of negative emotionality/low constraint on delinquency remains positive and significant (p<0.01) and the coefficient on the interaction effect is positive and significant (p<0.01) – individuals are score higher on the negative emotionality/low constraint scale are more likely to react to strain with increased delinquency.


Table A14. The Effect of Self-Control on Crime

Author(s) Main Variable(s) Data and Methods Causal Evidence Main Result(s)

Pratt and Cullen [2000]

Outcome(s): crime and analogous behaviors – measured in various ways in different studies Explanatory Variable(s): self control -- using Grasmick et al. (1993) inventory, other inventories, or behavioral questions

Data: Meta-analysis of 21 empirical studies testing the effects of self-control on crime Methods: Meta-analysis (mean effect sizes calculated)

Controls: Varies by study Timing of Measurements: Majority are contemporaneous, only two studies are longitudinal and thus have measures at different times Theory: Gottfredson and Hirschi (1990) General Theory of Crime. A uni-dimensional factor they label as “self-control” is responsible for much of the variance in crime and deviance across individuals

Mean effect size (standardized correlation coefficient) of self control on crime is 0.20 across the studies. When “behavioral” rather than “cognitive” or “attidunal” measures (i.e. Grasmick at al scale) are used the effect sizes are slightly higher. The effect of low self-control is significantly weaker in longitudinal studies as compared to cross-sectional studies. Self control has consistently large effects (minimum of 0.155) across different types of samples (juveniles vs. adults, racially homogenous or diverse, etc...). Variables that measure other criminal theories (i.e. social learning) do not effect the size of the effect of self-control but do tend to enter significantly in regressions on crime.

Nagin and Paternoster [1993]

Outcome(s): criminal proclivity -- students presented with crime scenarios were asked about whether they would participate and their probability of being caught if they do Explanatory Variable(s): self control -- using Grasmick et al. (1993) inventory

Data: Collected by authors; 699 undergraduates from the University of Maryland Methods: Tobit

Controls: gender, prior criminal behavior, sanctions present in the scenario, perceived utility from crime Timing of Measurements: The measures are contemporaneous Theory: Gottfredson and Hirschi (1990) General Theory of Crime. A uni-dimensional factor they label as “self-control” is responsible for much of the variance in crime and deviance across individuals

“Lack of self-control” as measured by the Grasmick et al. (1993) inventory is positively associated with the choice to commit crimes in all three scenarios (theft, drunk driving and sexual assault). Tobit regression was used because the modal response category for the dependent variables was 0 (i.e. no chance they would commit the crime). The lack of self-control measure was created by summing across 24 responses. An increase in 1 additional response associated with lack of self-control increases probability of intending to commit theft by 0.08 (p<0.01), drunk driving by 0.06 (p<0.01) and sexual assault by 0.11 (p<0.01).


Benda [2005]

Outcome(s): Crime -- self-reported property and person and offenses Explanatory Variable(s): self control -- using Grasmick et al. (1993) inventory and behavioral self-control scale where yes-no answers to questions about behavior (i.e. I regularly drive without a seatbelt) were combined to form an index

Data: Collected by authors; 3395 adolescents from a Midwestern state Methods: OLS

Controls: age, gender, race, rural/urban, family structure (two caregivers or not), caregiver years of education, annual family income, caregiver monitoring, Inventory of Parent and Peer Attachment (IPPA), Self-Efficacy Scale, 3 subscales from Childhood Trauma Questionnaire (CTQ), subscales from Multiple Problem Screening Inventory (MPSI) Timing of Measurements: Self-reports contemporaneous with personality measures but pertaining to offenses in the past Theory: Gottfredson and Hirschi (1990) General Theory of Crime. A uni-dimensional factor they label as “self-control” is responsible for much of the variance in crime and deviance across individuals

In an OLS regression of person and property offenses low behavioral self control is positively associated with both measures (p<0.01). These associations remain positive but shrink when the “cognitive” (Grasmick et al. (1993)) measure of self control is used instead.

Vazsonyi, Pickering, Junger et al. [2001]

Outcome(s): Deviance -- 55-item Normative Deviance Scale (NDS) - which contains 7 subscales relating to self-reported: vandalism, alcohol, drugs, school misconduct, general deviance, theft, and assault) Explanatory Variable(s): self control -- Grasmick et al. (1993) low self control scale

Data: International Study of Adolescent Development (ISAD), ~8500 subjects from Hungary, the Netherlands, Switzerland, and the United States. 6,085 15-19 year olds with no missing data were used in this present study Methods: Correlations, Hierarchical Regression Analysis

Controls: Sex, age, country Timing of Measurements: Contemporaneous self-repots about past deviance and self-control test. Reports given at ages 15-19 Theory: Gottfredson and Hirschi (1990) General Theory of Crime. A uni-dimensional factor they label as “self-control” is responsible for much of the variance in crime and deviance across individuals

All six sub-factors of self-control: impulsiveness, simple tasks, risk seeking, physical activity, self-centeredness, and temper were all positively correlated with deviance across all 7 deviance subscales. Risk seeking had the highest correlation with an average correlation of 0.320 across the 7 deviance subscales. Hierarchical regression was then used with each sub-factor of self-control entered in reverse order of correlation with deviance (age, sex, and country were entered first). Self control accounted for 18-24% of the variance in deviance across the 5 age groups (15-, 16-, 17-, 18-,19- year olds).Low self-control explains 10% of variance in theft, 12% in assault, 13% in alcohol use, 13% in drug use, 14% in school misbehavior, 15% in vandalism, and 16% in general deviance.


Table A15. The Effect of Education on Crime


Fergusson, John Horwood and Ridder [2005]

Outcome(s): crime -- self reported criminal activity, self-reported arrests and convictions, and self-reported incarcerations Explanatory Variable(s): IQ -- Revised Wechsler Intelligence Scale for Chidren (WISC-R)

Data: Christchurch Health and Development (CHDS) study - 1,265 children in a birth cohort in New Zealand Methods: OLS

Controls: Conduct problems age 7-9, attentional problems age 7-9, anxiety/withdrawal score (age 7-9), socioeconomic disadvantage score, family instability score, parental adjustment problems score, child abuse score, and gender Timing: IQ test at age 8-9, crime outcomes from ages 17-25 Theory: Though some have found a relationship between IQ and crime, IQ is also linked to childhood behavioral and conduct problems which may explain the link.

IQ was measured in 5 categories ranging from <85 to >115, with 10 IQ point contained within each of the intervening categories. An increase from 1 category to the next is associated with -.28 self-reported offenses committed (p<0.001) and -0.23 self reported arrests/convictions. However, once covariates are added to the regression this relationship goes away almost completely, with a coefficient of 0.00(p=0.87) on self-reported offenses and -0.11(p=0.11) on arrests/convictions.

Lynam, Moffitt and Stouthamer-Loeber [1993]

Outcome(s): Delinquency -- self-reports, teacher and parental reports Explanatory Variable(s): IQ - Short form WISC-R

Data: 430 12- and 13- year old boys from the Pittsburgh Youth Survey (PYS) Methods: ANOVA, mean differences

Controls: Analysis separately by age and race, social class as a covariate, effort during test Timing of Measurements: Reports were contemporaneous with test and were about delinquency at age 12 and 13. Theory:

The authors find a significant main effect of delinquency on IQ (p<0.001). Adding in social class does not change the differences in mean IQ scores of delinquents and nondelinquents.


Lochner and Moretti [2004]

Outcome(s): Imprisonment -- reported as institutionalized in the US census Crime (census) self-reported criminal acts (NLSY) Explanatory Variable(s): Education--years of schooling

Data: 1960, 1970 and 1980 US Census Data on males age 20-60. National Longitudnal Survey of Youth (NLSY) 1979 cohort. Methods: OLS, IV

Controls: Census - separate regressions for race, cohort birth effects, state of residents x year effects, age, year, state of birth, state of residence NLSY - Age/cohort, area of residence, dummy for school enrollment, family background, AFQT score, SMSA status, local unemployment rate. Timing of Measurements: Theory: Education may have an important impact on criminal outcomes through several mechanisms: education increases wage which increases opportunity cost of crime, education may change non-cognitive skills which may in turn impact criminal activity.

Census Data - Using compulsory schooling laws as an instrument in a 2SLS estimate they find that one additional year of schooling reduces the probability of being in prison at the time of the census by 0.1 (p<0.01) percentage points for whites and 0.3-0.5 (p<0.01) percentage points for blacks (depending on controls included). OLS estimates were remarkably similar. NLSY data - Self-reported crime participation is reduced by around 1-3 percentage points for each additional year of schooling for white males.. For black males the effect is not significant (the purport this to be due to underreporting of criminal activity by black males.


Machin, Marie and Vujic [2010]

Outcome(s): Crime -- apprehended or charged for a crime; self-reports of crime Imprisonment -- individuals in a prison service establishment during the British Census Explanatory Variable(s): Education--years of schooling

Data: UK Census 2001; British Crime Survey (BCS) from 2001/2 to 2007/8. Offender Index Database aggregated and combined with education information from Labour Force Survey and wages from New Earnings Survey 1984-2002. Methods: OLS, IV

Controls: Age dummies, county of brith dummies, gender dummies, non-white dummy, marital status, dummy for never worked, current country of residence Timing of Measurements: Theory: Education may have an important impact on criminal outcomes through several mechanisms: education increases wage which increases opportunity cost of crime, education may change non-cognitive skills which may in turn impact criminal activity.

Census Data - From an OLS regression - individuals with no qualifications are four times more likely to be in prison at the time of the census as those with some qualifications (p<0.01). BCS data - From an OLS regression - Individuals with no qualifications report significantly more self-reported crimes than those with qualifications. OID Cohort Data - Using changes in compulsory schooling laws as an IV- these 2SLS results are larger in magnitude than the OLS results but cannot be rejected as significantly different from OLS.


Table A16. The Effect of Personality and Preferences on Other Outcomes


Jaeger, Dohmen, Falk et al. [2010]

Outcome(s): migration – whether a person ever moved between regions in German between 2000 and 2006 Explanatory Variable(s): risk preference – the response to a survey question about general willingness to take risks, measured on a 10-point scale

Data: German Socio-Economic Panel (SOEP); 10,115 people aged 18-65 living in Germany (2000-2006) Methods: probit

Controls: (1) none (2) age and sex (3) age, sex, marital status, education, and place of origin Timing of Measurements: The risk question was asked in 2004 and 2006. A comparison of people who moved before and after 2004 suggested that migration did not affect risk preference. Theory: Migration is associated with uncertainty, so risk-preference should partially determine the propensity to move.

A one standard deviation increase in willingness to take risks is associated with a (1) 1.7 (p<0.01), (2) 1.1 (p<0.01), (3) 0.7 (p<0.01) percentage point increase in the probability of migrating, depending on specification (see controls). The average unconditional propensity was 5.8 percent.

Lundberg [2010]

Outcome(s): marital status – ever married by age 35, whether the first marriage ended in a divorce Explanatory Variable(s): personality – survey measures of The Big Five

Data: German Socio-Economic Panel (SOEP); 7,106 household heads, spouses, and partners aged 35-59 in Germany (2005) Methods: probit, Cox proportional hazard model

Controls: (1)education, German ethnicity, and living in East Germany (2) controls in (1), trust, risk aversion, locus of control, positive and negative reciprocity Timing of Measurements: The measures are contemporaneous. Theory: People match both positively and negatively on personality traits depending on whether they reflect preferences or specialization in production. Matching based on specialization will change as the labor market evolves.

Born before 1960: Extraversion increases the probability of marriage for both men and women (p<0.05); conscientiousness increases the probability for men (p<0.05); neuroticism increases the probability for women (p<0.05, (1) only); and agreeableness increases the probability for women (p<0.05; (1) only) but decreases the probability for men (p<0.05). Born after 1960: Openness to experience has a negative effect on marriage probabilities for men (p<0.05;(1)) and (p<0.10,(2)) and women (p<0.05;(2)) and conscientiousness has a strong positive effect for men and women (p<0.05;(2)).

A7.B. The Effects of Personality and Cognitive Measures...

Documents

Transcript of A7.B. The Effects of Personality and Cognitive Measures...